A method of compiling code that includes partitioning instructions in the code among a plurality of processors based on memory access latency associated with the instructions is disclosed. According to one aspect of the invention, partitioning instructions includes partitioning memory access dependence chains. Other embodiments are described and claimed.
|
1. A method of compiling code, comprising:
partitioning instructions in the code among a plurality of processors based on memory access latency associated with the instructions by:
partitioning memory access dependence chains into an upstream stage by assigning a first number of desired upstream nodes to the upstream stage, and also assigning instructions in the code on which the first number of desired upstream nodes are dependent to the upstream stage, wherein the first number of desired upstream nodes is N/d where N is a length of the memory access dependence chain and d is a pipelining degree; and
partitioning the memory access dependence chains into a downstream stage by assigning a last number of desired downstream nodes to the downstream stage, and assigning instructions in the code which are dependent on the last number of desired downstream nodes to the downstream stage, wherein the last number of desired downstream nodes is N*(d−1)/d;
performing the partitioning a plurality of times with subsequent partitioning being performed on the instructions assigned to the downstream stage.
7. An article of manufacture comprising a non-transitory machine accessible medium including sequences of instructions, the sequences of instructions including instructions which when executed cause the machine to perform:
partitioning instructions in code among a plurality of processors based on memory access latency associated with the instructions by:
partitioning memory access dependence chains into an upstream stage by assigning a first number of desired upstream nodes to the upstream stage, and also assigning instructions in the code on which the first number of desired upstream nodes are dependent on to the upstream stage, wherein the first number of desired upstream nodes is N/d where N is a length of the memory access dependence chain and d is a pipelining degree; and
partitioning the memory access dependence chains into a downstream stage by assigning a last number of desired downstream nodes to the downstream stage, and assigning instructions in the code which are dependent on the last number of desired downstream nodes to the downstream stage, wherein the last number of desired downstream nodes is N*(d−1)/d;
performing the partitioning a plurality of times with subsequent partitioning being performed on the instructions assigned to the downstream stage.
9. A code analysis unit implemented on a processor, comprising:
a dependence information unit to identify dependencies between instructions in code; and
a code partitioning unit to partition instructions in the code into a plurality of pipeline stages to be executed by a plurality of processors based on memory access latency associated with the instructions by
partitioning memory access dependence chains into an upstream stage by assigning a first number of desired upstream nodes to the upstream stage, and also assigning instructions in the code on which the first number of desired upstream nodes are dependent to the upstream stage, wherein the first number of desired upstream nodes is N/d where N is a length of the memory access dependence chain and d is a pipelining degree; and
partitioning the memory access dependence chains into a downstream stage by assigning a last number of desired downstream nodes to the downstream stage, and assigning instructions in the code which are dependent on the last number of desired downstream nodes to the downstream stage, wherein the last number of desired downstream nodes is N*(d−1)/d;
performing the partitioning a plurality of times with subsequent partitioning being performed on the instructions assigned to the downstream stage.
2. The method of
3. The method of
if a weight required for execution of the upstream stage exceeds a predetermined value, re-computing the number of desired upstream nodes;
assigning the re-computed number of desired upstream nodes to the upstream stage; and
assigning instructions in the code on which the re-computed number of desired upstream nodes are dependent to the upstream stage.
4. The method of
5. The method of
6. The method of
constructing a memory access dependence graph; and
identifying a memory access dependence chain from the memory access dependence graph.
8. The article of manufacture of
10. The apparatus of
an evaluation unit to determine whether a weight of the upstream stage exceeds a predetermined value.
11. The apparatus of
12. The apparatus of
13. The apparatus of
14. The apparatus of
15. The apparatus of
|
|||||||||||||||||||||||||
An embodiment of the present invention relates to compilers. More specifically, an embodiment of the present invention relates to a method and apparatus for partitioning programs to balance memory latency.
Processor designs are moving towards multiple core architectures where more than one core (processor) is implemented on a single chip. Multiple core architectures provide increased computing power while requiring less space and a lower amount of power. Multiple core architectures are particularly useful for pipelining instructions in applications that require high processing speeds, such as packet processing in networks which may require processing speeds of up to 10 Gigabits per second. The instructions may be pipelined, for example, into stages where each stage is supported by a different processor or processor core.
The performance of pipelined computations as a whole can be no faster than the slowest of the pipeline stages. For this reason, when pipelining instructions, compilers attempt to balance instructions among stages as evenly as possible. It is common for compilers to partition instructions between stages based upon the compute cycles required for executing instructions. This technique may be effective in some instances. However, when the instructions include a large number of memory accesses, the latency required for completing some memory accesses may produce additional undesired delay that is not accounted for by the compilers. For example, while the latency of two independent memory accesses may be overlapped with each other, instructions that depend on the completion of a particular memory access operation cannot be executed until the memory access is completed. Hence, instructions with dependencies on memory access operations cannot be overlapped with the latency of the memory access.
Thus, what is needed is a method and apparatus for partitioning programming to balance memory latency.
The features and advantages of embodiments of the present invention are illustrated by way of example and are not intended to limit the scope of the embodiments of the present invention to the particular embodiments shown.
In the following description, for purposes of explanation, specific nomenclature is set forth to provide a thorough understanding of embodiments of the present invention. However, it will be apparent to one skilled in the art that specific details in the description may not be required to practice the embodiments of the present invention. In other instances, well-known components, programs, and procedures are shown in block diagram form to avoid obscuring embodiments of the present invention unnecessarily.
The memory 113 may be a dynamic random access memory device, a static random access memory device, read-only memory, and/or other memory device. The memory 113 may store instructions and code represented by data signals that may be executed by the processor 101. According to an example embodiment of the computer system 100, a compiler may reside in a different computer system and generate destination instruction codes which are downloaded and executed on the computer system 100. Alternatively the compiler may be stored in the memory 113 and implemented by the processors 101 and 102 in the computer system 100. The compiler may partition programs to balance memory latency. According to one embodiment, the compiler partitions instructions in the code of a program among the processors 101 and 102 based on memory access latency associated with the instructions.
A cache memory may reside inside each of the processors 101 and 102 to store data signals stored in memory 113. The cache speeds access to memory by the processor 101 by taking advantage of its locality of access. In an alternate embodiment of the computer system 100, the cache resides external to the processor 101. A bridge/memory controller 111 is coupled to the CPU bus 110 and the memory 113. The bridge/memory controller 111 directs data signals between the processor 101, the memory 113, and other components in the computer system 100 and bridges the data signals between the CPU bus 110, the memory 113, and a first input output (IO) bus 120.
The first IO bus 120 may be a single bus or a combination of multiple buses. The first IO bus 120 provides communication links between components in the computer system 100. A network controller 121 is coupled to the first IO bus 120. The network controller 121 may link the computer system 100 to a network of computers (not shown) and supports communication among the machines. A display device controller 122 is coupled to the first IO bus 120. The display device controller 122 allows coupling of a display device (not shown) to the computer system 100 and acts as an interface between the display device and the computer system 100.
A second IO bus 130 may be a single bus or a combination of multiple buses. The second IO bus 130 provides communication links between components in the computer system 100. A data storage 131 is coupled to the second IO bus 130. The data storage 131 may be a hard disk drive, a floppy disk drive, a CD-ROM device, a flash memory device or other mass storage device. An input interface 132 is coupled to the second IO bus 130. The input interface 132 may be, for example, a keyboard and/or mouse controller or other input interface. The input interface 132 may be a dedicated device or can reside in another device such as a bus controller or other controller. The input interface 132 allows coupling of an input device to the computer system 100 and transmits data signals from an input device to the computer system 100. An audio controller 133 is coupled to the second IO bus 130. The audio controller 133 operates to coordinate the recording and playing of sounds and is also coupled to the IO bus 130. A bus bridge 123 couples the first IO bus 120 to the second IO bus 130. The bus bridge 123 operates to buffer and bridge data signals between the first IO bus 120 and the second IO bus 130.
The compiler 200 includes a front end unit 220. According to an embodiment of the compiler 200, the front end unit 220 operates to parse source code and convert it to an abstract syntax tree.
The compiler 200 includes an intermediate language (IL) unit 230. The intermediate language unit 230 transforms the abstract syntax tree into a common intermediate form such as an intermediate representation. It should be appreciated that the intermediate language unit 230 may transform the abstract syntax tree into one or more common intermediate forms.
The compiler 200 includes a code analysis unit 240. The code analysis unit 240 includes a dependence information unit 241. According to an embodiment of the code analysis unit 240, the dependence information unit 241 identifies instruction dependence information such as flow dependence and control dependence between instructions in the code. The dependence information unit 241 may generate a memory access dependence graph and memory access dependence chains from the instruction dependence information. The code analysis unit 240 includes a code partitioning unit 242 that partitions instructions in the code among a plurality of processors based on memory access latency associated with the instructions.
The compiler 200 includes a code generator unit 250. The code generator unit 250 converts the intermediate representation into machine or assembly code.
The code partitioning unit 300 includes a length unit 320. The length unit determines a number of nodes from a memory access dependence chain to allocate to an upstream stage and a downstream stage. The nodes from the memory access dependence chain represent memory access instructions. The upstream stage may be designated as a pipeline stage. The downstream stage may be designated as one or more pipeline stages after the upstream stage. According to an embodiment of the present invention, the number desired upstream nodes to allocate to the upstream stage is N/d and the number of desired downstream nodes to allocate to the downstream stage is N*(d−1)/d, where N is the length of the memory access dependence chain, and d is the pipelining degree. It should be appreciated that this relationship may be adjusted according to the actual compute environment.
The code partitioning unit 300 includes an assignment unit 330. The assignment unit 330 assigns a first number of desired upstream nodes in a memory access dependence chain to the upstream stage. The assignment unit 330 may also assign a last number of desired downstream nodes in a memory access dependence chain to the downstream stage.
The code partitioning unit 300 includes a close up unit 340. The close up unit 340 assigns instructions in the code which may include memory access or non-memory access instructions for which the first number of desired upstream nodes is dependent on to the upstream stage. After the memory access dependence chain has been processed for downstream stage assignment, the close up unit 340 may also assign instructions in the code which may include memory access or non-memory access instructions for which depends on the last number of desired downstream nodes to the downstream stage. The close up unit 340 assigns an instruction only once. Only remaining unassigned instructions are available for assignment to other stages.
The code partitioning unit 300 includes an evaluation unit 350. The evaluation unit 350 determines whether a computed weight for executing the instructions assigned to the upstage stream or the downstream stage exceeds a predetermined value. If the computed weight required for executing the instructions in the upstream stage or the downstream stage exceeds a predetermined value, a new number of desired upstream nodes or a new number of desired downstream nodes is determined. According to an embodiment of the code partitioning unit 300, the new number may include one less number of desired upstream nodes or number of desired downstream nodes.
The code partitioning unit 300 includes a balancing unit 360. According to an embodiment of the present invention, the components of the code partitioning unit 300 process each memory access dependence chain to assign instructions to either an upstream stage or a downstream stage. After all the memory access dependence chains have been processed, the balancing unit 360 assigns the remaining unassigned instructions in the code to either the upstream stage or downstream stage. The code partitioning unit 300 partitions instructions to two stages at a time. The instructions assigned to an upstream stage represent instructions that may be assigned to a single pipelined stage. The instructions assigned to the downstream stage may require further partitioning by the code partitioning unit 300 in order to identify instructions for additional pipelined stages if the pipelining degree (number of pipelined stages) is greater than 2.
At 402, a memory access dependence graph is generated. The memory access dependence graph may be generated from the instruction independence information determined at 401.
At 403, memory access dependence chains are generated from the memory access dependence graph generated at 402. According to an embodiment of the present invention, a memory access dependence chain is a path (n1, n2, . . . nk) in the memory access dependence graph where n1 has no predecessors and nk has no successor, and can be computed by traversing the memory access dependence graph.
At 404, the memory access dependence chains are partitioned. The memory access dependence chains are partitioned using a set of procedures where the instructions in the code are partitioned into two stages each time the set of procedures is performed. For d-way pipelining transformation, where d is the pipelining degree, the set of procedures is performed d−1 times. The set of procedures partitions the instructions in the code to an upstream stage and a downstream stage. The set of procedures may be subsequently performed on the instructions assigned to the downstream stage when necessary. According to an embodiment of the present invention, the memory access dependence chains are partitioned in a decreasing order of length.
At 405, the remaining instructions are assigned. According to an embodiment of the present invention, the remaining instructions may be assigned while trying to balance a computed weight among the pipelined stages, or using other techniques.
At 502, control terminates the procedure.
At 503, a work queue is emptied.
At 504, all instructions that are dependent on memory access instruction M are represented as nodes and are placed in the work queue.
At 505, it is determined whether there is a node, N, in the work queue. If there is not a node in the work queue, control returns to 501. If there is a node in the work queue, control proceeds to 506.
At 506, node N is removed from the work queue.
At 507, it is determined whether node N is a memory access instruction. If it is determined that node N is a memory access instruction, control proceeds to 508. If it is determined that node N is not a memory access instruction, control proceeds to 509.
At 508, memory access dependence M is connected to N in the memory access dependence graph.
At 509, all instructions that are dependent on the instruction represented by node N and have not been visited before are placed into the work queue. Control returns to 505.
At 602, a first number of desired upstream nodes in a memory access dependence chain are assigned to the upstream stage.
At 603, the instructions in the upstream stage are closed up. According to an embodiment of the present invention, closing up includes assigning instructions in the code, which may include non-memory access instructions, for which the first number of desired upstream nodes are dependent on to the upstream stage.
At 604, it is determined whether a computed weight for executing the instructions assigned to the upstage stream exceeds a predetermined value. If the computed weight required for executing the instructions in the upstream stage exceeds a predetermined value, control proceeds to 605. If the computed weight required for executing the instructions in the upstream stage does not exceed the predetermined value, control proceeds to 606.
At 605, a new number of desired upstream nodes to allocate to the upstream stage is determined. According to an embodiment of the present invention, the new number may be the previous number subtracted by one.
At 606, the assignments made to the upstream stage are utilized.
Referring to
At 702, a last number of desired downstream nodes in a memory access dependence chain are assigned to the downstream stage.
At 703, the instructions in the downstream stage are closed up. According to an embodiment of the present invention, closing up may include assigning instructions in the code, which may include memory and non-memory access instructions, which depend on the last number of desired downstream nodes to the downstream stage.
At 704, it is determined whether a computed weight for executing the instructions assigned to the downstage stream exceeds a predetermined value. If the computed weight required for executing the instructions in the downstream stage exceeds a predetermined value, control proceeds to 705. If the computed weight required for executing the instructions in the downstream stage does not exceed the predetermined value, control proceeds to 706.
At 705, a new number of desired downstream nodes to allocate to the downstream stage is determined. According to an embodiment of the present invention, the new number may be the previous number subtracted by one.
At 706, the assignments made to the downstream stage are utilized.
According to an embodiment of the present invention, memory access instructions are allocated among program partitions in the pipelining transformation of applications. The memory access latency in each pipeline stage is effectively hidden by overlapping the latency of memory accesses and other operations. This is achieved by summarizing the dependence between the memory access instructions in the program, constructing dependence chains of the memory access instructions, and partitioning the memory access dependence chains evenly among the pipeline stages.
A group of exemplary instructions which may be partitioned across two pipelined stages according to an embodiment of the present invention is shown below.
(1)
t1 = f( );
(2)
t2 = read(t1)
(3)
t3 = read (t1 + 1)
(4)
g(t3)
(5)
t4 = t2 +c1
(6)
t5 = read (t4)
(7)
h(t5)
(8)
if (t3>0)
{
(9)
k(t5)
(10)
t6 = read (t5)
(11)
t7 = t4 +1
(12)
write (t7, t6)
}
Referring to
At 402, a memory access dependence graph is generated. The memory access dependence graph may be generated from the instruction independence information. The memory access dependence graph shown in
At 403, memory access dependence chains are generated from the memory access dependence graph. The memory access dependence chains shown in
At 404, the memory access dependence chains are partitioned. The set of procedures shown in
Referring to
At 602, the first 2 nodes, nodes 2 and 6 that correspond to instructions (2) and (6) are assigned to the upstream stage.
At 603, instructions (2) and (6) are dependent on instructions (1) and (5), thus nodes 1 and 5 are also assigned to the upstream stage.
Assuming that the computed weight of instructions (1), (2), (5), and (6) does not exceed a predetermined value, the assignments made at 602 and 603 are utilized.
Referring to
At 702, the last 2 nodes, nodes 10 and 12 that correspond to instructions (10) and (12) are assigned to the downstream stage.
At 703, since no nodes are dependent on instructions (10) and (12) which have not already been considered, no additional nodes are assigned to the downstream stage.
Referring back to
At 602, the first 2 nodes, nodes 2 and 6 have already been assigned to the upstream stage. Thus, no further assignment of nodes in the memory access dependence chain or other nodes are assigned to the upstream stage. Control proceeds to
At 701, the DesiredLengthofDownstream=N*(d−1)/d=3*(2−1)/2=1.5, which rounds to 2.
At 702, the last 2 nodes, nodes 6 and 12 have already been assigned to the downstream stage. Thus, no further assignment of nodes in the memory access dependence chain or other nodes are assigned to the downstream stage.
Referring back to
At 602, of the first 2 nodes, nodes 3 and 10, node 3, which corresponds to instruction (3) is assigned to the upstream stage. Node 10 has already been assigned to the downstream stage.
At 603, instruction (3) is dependent on instruction (1) which has already been assigned to the upstream stage.
Assuming that the computed weight of instructions (1), (2), (3), (5), and (6) does not exceed a predetermined value, the new assignments made at 602 is utilized.
Referring to
At 702, the last 2 nodes, nodes 10 and 12 have already been assigned to the downstream stage. Thus, no further assignment of nodes in the memory access dependence chain or other nodes are assigned to the downstream stage.
Referring back to
At 602, the first nodes, node 3, has already been assigned to the upstream stage. Thus, no further assignment of nodes in the memory access dependence chain or other nodes are assigned to the upstream stage. Control proceeds to
At 701, the DesiredLengthofDownstream=N*(d−1)/d=2*(2−1)/2=1.
At 702, the last node, nodes 12 has already been assigned to the downstream stage. Thus, no further assignment of nodes in the memory access dependence chain or other nodes are assigned to the downstream stage.
Thus, nodes (1), (2), (3), (5), and (6) are assigned to the upstream stage, and nodes (10) and (12) are assigned to the downstream stage.
Referring back to
Embodiments of the present invention may be provided as a computer program product, or software, that may include a machine-readable medium having stored thereon instructions. The machine-readable medium may be used to program a computer system or other electronic device. The machine-readable medium may include, but is not limited to, floppy diskettes, optical disks, CD-ROMs, and magneto-optical disks or other type of media/machine-readable medium suitable for storing electronic instructions. The techniques described herein are not limited to any particular software configuration. They may find applicability in any computing or processing environment. The term “machine readable medium” used herein shall include any medium that is capable of storing or encoding a sequence of instructions for execution by the machine and that cause the machine to perform any one of the methods described herein. Furthermore, it is common in the art to speak of software, in one form or another (e.g., program, procedure, process, application, module, unit, logic, and so on) as taking an action or causing a result. Such expressions are merely a shorthand way of stating that the execution of the software by a processing system causes the processor to perform an action to produce a result.
In the foregoing specification embodiments of the invention has been described with reference to specific exemplary embodiments thereof. It will, however, be evident that various modifications and changes may be made thereto without departing from the broader spirit and scope of the embodiments of the invention. The specification and drawings are, accordingly, to be regarded in an illustrative rather than restrictive sense.
| Patent | Priority | Assignee | Title |
| 10372428, | Jun 16 2015 | ARCHITECTURE TECHNOLOGY CORPORATION | Dynamic computational acceleration using a heterogeneous hardware infrastructure |
| 10558441, | Mar 12 2018 | International Business Machines Corporation | Compiler for restructuring code using iteration-point algebraic difference analysis |
| 10942716, | Jun 16 2015 | ARCHITECTURE TECHNOLOGY CORPORATION | Dynamic computational acceleration using a heterogeneous hardware infrastructure |
| 9317595, | Dec 06 2010 | YAHOO ASSETS LLC | Fast title/summary extraction from long descriptions |
| 9501304, | Jun 16 2015 | ARCHITECTURE TECHNOLOGY CORPORATION | Lightweight application virtualization architecture |
| 9983857, | Jun 16 2015 | ARCHITECTURE TECHNOLOGY CORPORATION | Dynamic computational acceleration using a heterogeneous hardware infrastructure |
| Patent | Priority | Assignee | Title |
| 5179699, | Jan 13 1989 | INTERNATIONAL BUSINESS MACHINES CORPORATION, ARMONK, NY 10504, A CORP OF NY | Partitioning of sorted lists for multiprocessors sort and merge |
| 5524264, | Mar 11 1993 | Sony Corporation | Parallel arithmetic-logic processing device |
| 5642512, | Sep 13 1994 | Matsushita Electric Co. | Compiler with improved live range interference investigation |
| 5768594, | Jul 14 1995 | THE CHASE MANHATTAN BANK, AS COLLATERAL AGENT | Methods and means for scheduling parallel processors |
| 5872972, | Jul 05 1996 | TERADATA US, INC | Method for load balancing a per processor affinity scheduler wherein processes are strictly affinitized to processors and the migration of a process from an affinitized processor to another available processor is limited |
| 6374403, | Aug 20 1999 | Qualcomm Incorporated | Programmatic method for reducing cost of control in parallel processes |
| 6587866, | Jan 10 2000 | Oracle America, Inc | Method for distributing packets to server nodes using network client affinity and packet distribution table |
| 6601084, | Dec 19 1997 | Citrix Systems, Inc | Dynamic load balancer for multiple network servers |
| 6681388, | Oct 02 1998 | MINISTER OF ECONOMY, TRADE AND INDUSTRY | Method and compiler for rearranging array data into sub-arrays of consecutively-addressed elements for distribution processing |
| 6691306, | Dec 22 2000 | AVAGO TECHNOLOGIES INTERNATIONAL SALES PTE LIMITED | Use of limited program space of general purpose processor for unlimited sequence of translated instructions |
| 6728748, | Dec 01 1998 | Network Appliance, Inc | Method and apparatus for policy based class of service and adaptive service level management within the context of an internet and intranet |
| 6952816, | Oct 07 2002 | SAMSUNG ELECTRONICS CO , LTD | Methods and apparatus for digital circuit design generation |
| 6970929, | Jun 12 2002 | JUNGSEOK-INHA SCHOOL S FOUNDATION | Vector-based, clustering web geographic information system and control method thereof |
| 7103879, | Jan 16 2001 | Canon Kabushiki Kaisha | Method and device for partitioning a computer program |
| 7162579, | Aug 19 2002 | ADAPTEC INC | Asymmetrical load balancing for mirrored storage volumes |
| 7363467, | Jan 03 2002 | Intel Corporation | Dependence-chain processing using trace descriptors having dependency descriptors |
| 7406692, | Feb 24 2003 | Oracle International Corporation | System and method for server load balancing and server affinity |
| 7908355, | Jun 20 2002 | International Business Machines Corporation | Method for improving network server load balancing |
| 8176479, | Jul 21 2006 | National Instruments Corporation | Case structure replacement in a data flow diagram |
| 20020124240, | |||
| 20030037319, | |||
| 20030126408, | |||
| 20040158694, | |||
| 20050149940, | |||
| 20060026599, | |||
| 20060037024, | |||
| 20060112377, | |||
| 20090089765, | |||
| 20110067017, | |||
| EP363882, | |||
| JP4070439, |
| Executed on | Assignor | Assignee | Conveyance | Frame | Reel | Doc |
| Dec 17 2005 | Intel Corporation | (assignment on the face of the patent) | / | |||
| Feb 22 2006 | JIANG, XIAODAN | Intel Corporation | ASSIGNMENT OF ASSIGNORS INTEREST SEE DOCUMENT FOR DETAILS | 026586 | /0152 | |
| Feb 22 2006 | DAI, JINQUAN | Intel Corporation | ASSIGNMENT OF ASSIGNORS INTEREST SEE DOCUMENT FOR DETAILS | 026586 | /0152 |
| Date | Maintenance Fee Events |
| Aug 27 2013 | ASPN: Payor Number Assigned. |
| Mar 09 2017 | M1551: Payment of Maintenance Fee, 4th Year, Large Entity. |
| May 17 2021 | REM: Maintenance Fee Reminder Mailed. |
| Nov 01 2021 | EXP: Patent Expired for Failure to Pay Maintenance Fees. |
| Date | Maintenance Schedule |
| Sep 24 2016 | 4 years fee payment window open |
| Mar 24 2017 | 6 months grace period start (w surcharge) |
| Sep 24 2017 | patent expiry (for year 4) |
| Sep 24 2019 | 2 years to revive unintentionally abandoned end. (for year 4) |
| Sep 24 2020 | 8 years fee payment window open |
| Mar 24 2021 | 6 months grace period start (w surcharge) |
| Sep 24 2021 | patent expiry (for year 8) |
| Sep 24 2023 | 2 years to revive unintentionally abandoned end. (for year 8) |
| Sep 24 2024 | 12 years fee payment window open |
| Mar 24 2025 | 6 months grace period start (w surcharge) |
| Sep 24 2025 | patent expiry (for year 12) |
| Sep 24 2027 | 2 years to revive unintentionally abandoned end. (for year 12) |